Available online at www.sciencedirect.

com

ScienceDirect
Procedia Engineering 155 (2016) 313 – 322

International Symposium on "Novel structural skins - Improving sustainability and efficiency

through new structural textile materials and design

Beyond static analysis: Investigation of membrane structure

performance using time stepping, transient and progressive analyses

Adam Bowna*, Tom Makina, David Wakefielda

a
Tensys Ltd, 1 St Swithins Yard, Walcot Street, Bath, BA1 5BG, UK

Abstract

Due to recent improvements in commercially available programming and modelling tools, the role of the tension structures
engineer is increasingly being expanded beyond the traditional disciplines of form finding, static load analysis and creation of
fabrication geometry. The engineer is now able to consider a variety of peripheral analytical considerations, including transient or
time stepping studies. These developments may require the enhancement of in-house software tools or combining current
software capabilities with third party tools. This concept is illustrated with reference to three examples applied to recent
structures; the creation of a hydraulic flow tool allowing the assessment of water flow over a complex 3d surface, automated time
stepping of existing static analysis tools to provide an assessment of structural response under a transient hydrostatic load and
finally a batch processed sequence of static analyses to determine the risk and potential mechanisms for a progressive collapse.
© 2016 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
© 2016 The Authors. Published by Elsevier Ltd.
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of the TensiNet Association and the Cost Action TU1303, Vrije Universiteit Brussel.
Peer-review under responsibility of the TensiNet Association and the Cost Action TU1303, Vrije Universiteit Brussel
Keywords: tension structure, analysis, hydraulic flow, time stepping, transient analysis, ponding, progressive collapse, batch analysis

* Corresponding Author. Tel.:+44 (0)1225 321950 fax: +44 (0)1225 321969

E-mail address: adam.bown@tensys.com

1877-7058 © 2016 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of the TensiNet Association and the Cost Action TU1303, Vrije Universiteit Brussel
doi:10.1016/j.proeng.2016.08.034
314 Adam Bown et al. / Procedia Engineering 155 (2016) 313 – 322

1. Introduction

1.1. Background

The tension structures engineer has been undertaking static analyses of structures with finite element models
using commercial or in-house software codes for many years. Full dynamic, transient and time stepping analysis is
also facilitated in some of the larger commercial codes, although this option often comes at significant expense and
the inner workings can often appear opaque to non-regular users. If the engineer is required extend their capabilities,
it may be preferable to look for ways to build on existing code or combine with other commercial software tools.
This paper presents three examples where an existing static analysis code has been enhanced, combined with
other software or used in a batch processing environment, to increase analytical capabilities. Namely:

• A simplified hydraulic flow tool to determine unsteady flow patterns over any irregular surface.
• A transient time stepping tool to determine the behaviour of a water pond occurring mid structure.
• A transient propagation analysis to determine the risk of total failure of a tension structure upon failure or
removal of a key element.

Conclusions for each example will be presented at the end of the relevant section.

1.2. Software

The three examples discussed within this paper have are founded upon inTENS, the Tensys Ltd in-house software
suite for tension structures. inTENS uses a dynamic relaxation solver with kinetic damping control which readily
handles large deflections – a key requirement for the nature of the problems discussed herein [1,2]. inTENS uses
ASCII files for both input and output and the key analytical programs are invoked from a Windows command line;
both these points mean the code is well suited to automated batch processing.
Much of the coding to facilitate these examples is undertaken and controlled within Rhino 3D, using a custom
modeling plugin referred to as ‘RhinoTools’.

2. Hydraulic Flow Modeling

2.1. Introduction

The Tensys Hydro package simulates unsteady fluid flow across an arbitrary mesh surface. The application solves
depth-averaged free surface flow equations to obtain water depth, and flow rate across the analysis domain. It allows
assessment of flow patterns, drainage requirements, ponding locations and the effect of local surface obstructions.

2.2. Key Assumptions and Equations

The solver applies the 2D shallow water equations on a regular computational grid. These equations are solved
iteratively by considering mass and momentum flux across the boundaries of each grid cell. The following
assumptions are made when considering the flow analysis:

• Viscous forces are negligible

• Depth averaged flow
• Only horizontal velocity components propagated between cells.

Shallow water theory assumes that the horizontal dimensions are significantly greater than the vertical. The
accuracy of the model thus decreases for very steep surface gradients or rapidly changing flow conditions.
Given the assumptions above, the shallow water equations can be written as:
 Adam Bown et al. / Procedia Engineering 155 (2016) 313 – 322 315

݄௧ ൅ ሺ݄‫ݑ‬ሻ௫ ൅ ሺ݄‫ݒ‬ሻ௬ ܴሺ‫ݔ‬ǡ ‫ݕ‬ǡ ‫ݐ‬ሻ

௚ ଶ
‫ݕݐ݅ݑ݊݅ݐ݊݋ܥ‬
ଶ
൦ ሺ݄‫ݑ‬ሻ ௧ ൅ ሺ݄‫ݑ‬ ൅ ଶ
݄ ሻ௫ ൅ ሺ݄‫ݒݑ‬ሻ ௬ ൪ൌ቎ െ݄݃‫ܤ‬ ௫
௫ െ ߬ ቏ ൌ  ൥‫ ܺ݉ݑݐ݊݁݉݋ܯ‬൩
௚
ሺ݄‫ݑ‬ሻ௧ ൅ ሺ݄‫ݒݑ‬ሻ௫ ൅  ሺ݄‫ ݒ‬ଶ ൅ ݄ଶ ሻ௬ െ݄݃‫ܤ‬௬ െ ߬ ௬ ‫ܻ݉ݑݐ݊݁݉݋ܯ‬
ଶ
Where:
‫ = ݐ‬time
‫ = ݔ‬x-direction
‫ = ݕ‬y-direction
݄ = water depth
‫ = ܤ‬bed topography
ܴሺ‫ݔ‬ǡ ‫ݕ‬ǡ ‫ݐ‬ሻ is the source term, such as rainfall in m/s
߬ ௫ and ߬ ௬ are the frictional accelerations per unit width

డ௛
The subscripts indicate the partial derivative with respect the given quantity. i.e. ݄௧ ൌ 
డ௧

The െ݄݃‫ܤ‬௫ and െ݄݃‫ܤ‬௫ terms represent the gravitational acceleration in the x and y directions. For steep slopes
these terms can become unrealistically large. Solution of these equations requires a specialized multistep Runge-
Kutta solver to maintain numerical stability for low water depths. The solution technique is described in detail by
Chertock et Al [3].
In the steady state what comes in must come out again, and thus shallow water theory can be used to assess flow
rates at outflow drainage points under constant rain loading with a high degree of confidence. However, the
determination of local water depth is more error prone as it is very sensitive to the chosen friction factor. Without
careful calibration against experimental testing for a similar problem, the calculated water depths thus need to be
used with care.

2.3. Examples

The following figures provide examples of the Tensys Hydro tool applied to real structures.

Fig. 1. Khalifa Stadium – Visualization of flow paths across a multi-bay Fig. 2. Queensland Velodrome Roof – Identifying hotspots for
structure under constant rainfall loading peak flow under extreme rainfall even
316 Adam Bown et al. / Procedia Engineering 155 (2016) 313 – 322

3. Time Stepped Ponding Analysis

3.1. Introduction

The figure below shows a typical hydrostatic loading case. A dam (shown in red) is inadvertently introduced at
the edge of a standard barrel vaulted membrane structure allowing rainwater to back up thus applying hydrostatic
loading to the membrane surface. The user defined zero head datum for the hydrostatic load is fixed (shown in
purple) and the structure is readily analysed. The load applied to each membrane element is given by:

Le = Y·g·de Where: Le = The load applied to each element of the FE model

Y = density of water in kg/m3
g = gravitational acceleration in m/s2
de = depth of water above each triangular element in m

Datum point
(Fixed)

Fig. 3. Standard hydrostatic loading at the edge of a typical membrane structure caused by the introduction of a dam.

However, consider the situation of a ‘seed’ event such as drifted snow settling on the surface of the structure
causing a local depression within the body of a membrane structure. Assume the seed event remains in place and
there is a subsequent rainfall event such that rainwater is able to accumulate in the depression.
The zero datum for hydrostatic loading (lowest point on depression perimeter), may deflect during the course of
the run, whilst the user defined zero datum will typically remain constant. This may give rise to a situation where the
structure deflection leads to the applied hydrostatic loading increasing indefinitely during the course of the static
run. The datum point is required to track the low point of the pond, in order to update the applied load accordingly.
This section presents Tensys Pondr; an automated method for handling this situation using a time stepped
sequence of static cases, with reference to some examples applied to a pneumatic structure. This type of structure
has been chosen for two reasons; firstly because the problem becomes multi-functional, the propagation of the pond,
is closely related to the pressure control within the structure. Secondly because the ponding behaviour is potentially
more spectacular than a fixed boundary membrane structure.
The figure below illustrates a typical pneumatic structure and the formation of an initial pond following a seed
event of a weight settling on the structure surface.

Internal pressure Pi

Pond catchment area

Internal pressure Pi

Fig. 4. Seed event and subsequent rainfall causing hydrostatic loading onto the surface of a typical pneumatic structure.
 Adam Bown et al. / Procedia Engineering 155 (2016) 313 – 322 317

3.2. Batch Processing

^ƚĂƌƚŝŶŐĐŽŶĚŝƚŝŽŶ͗&ŵŽĚĞůŽĨƉŶĞƵŵĂƚŝĐƐƚƌƵĐƚƵƌĞƐƵƉƉŽƌƚĞĚďLJ
ƵŶŝĨŽƌŵŝŶƚĞƌŶĂůƉƌĞƐƐƵƌĞ͘^ƚƌƵĐƚƵƌĞŝŶƐƚĂƚŝĐĞƋƵŝůŝďƌŝƵŵ

^ĞĞĚĞǀĞŶƚĚĞĨŝŶĞĚ͗;ƐƵĐŚĂƐĚĂŵ͕ĚƌŝĨƚĞĚƐŶŽǁůŽĂĚ͕ĚĞďƌŝƐŽŶ
ƐƵƌĨĂĐĞ͕ƉŽƐƐŝďůLJĐŽƵƉůĞĚǁŝƚŚƉƌĞƐƐƵƌĞĐŽŶƚƌŽůŵĂůĨƵŶĐƚŝŽŶͿ
WZ^^/KE&KZD

ZĂŝŶĞǀĞŶƚŽĐĐƵƌƐ͘^ƉĞĐŝĨLJƌĂŝŶŝŶƚĞŶƐŝƚLJΖƌΖ;ŝŶŵŵͬŚƌĨŽƌĞdžĂŵƉůĞͿ
ĂŶĚŶŽƚŝŽŶĂůƚŝŵĞƐƚĞƉΖƚΖ

'ĞŽŵĞƚƌŝĐĂƐƐĞƐŵĞŶƚŽĨĚĞƉƌĞƐƐŝŽŶƚŽĐĂůĐƵůĂƚĞůŽǁĞƐƚĚĂƚƵŵƉŽŝŶƚ͕
ĐĂƚĐŚŵĞŶƚĂƌĞĂĂŶĚǀŽůƵŵĞĨŽƌĐƵƌƌĞŶƚƐƚĞƉΖŶΖ͕Ğ͘Ő͘njŶ͕Ŷ͕sŶ

'ĞŽŵĞƚƌŝĐĂƐƐĞƐŵĞŶƚŽĨƉŶĞƵŵĂƚŝĐƐƚƌƵĐƚƵƌĞǀŽůƵŵĞĂŶĚƉƌĞƐƵƌĞƚŽ
ĚĞƚĞƌŵŝŶĞWϬͼsϬ сĐŽŶƐƚĂŶƚ<

/ĨƌĞƋƵŝƌĞĚ͕ŝŶƚĞƌŶĂůƉƌĞƐƐƵƌĞĐŽƌƌĞĐƚĞĚĂĐĐŽƌĚŝŶŐƚŽŶĞǁǀŽůƵŵĞƚŽ
ŵĂŝŶƚĂŝŶWŶͼsŶсĐŽŶƐƚĂŶƚ<

&ŽƌŝŶŝƚŝĂůƐƚĞƉƚϬ ͕ĚĞƉƌĞƐƐŝŽŶΖĨŝůůĞĚΖǁŝƚŚǁĂƚĞƌƚŽŝŶŝƚŝĂůĚĂƚƵŵϬ͘&Žƌ
ƐƵďƐĞƋƵĞŶƚƐƚĞƉƐĂĚĚŝƚŝŽŶĂůǁĂƚĞƌůŽĂĚ>ĚĞůƚĂ ĐĂůĐƵůĂƚĞĚĨƌŽŵ;ƌͼƚͿĨŽƌĐƵƌƌĞŶƚ
ƚŝŵĞƉĞƌŝŽĚĂŶĚĐĂƚĐŚŵĞŶƚĂƌĞĂ͘dŽƚĂůůŽĂĚ>ŶĚĞĨŝŶĞĚďLJ>Ŷ с>ŶͲϭн>ĚĞůƚĂ
&ŽƌƚŚĞĐĂƐĞǁŚĞƌĞsŶ фsŶͲϭ͕Ğ͘Ő͘ƚŚĞĐĂƐĞŽĨƚŚĞĚĞƉƌĞƐƐŝŽŶƌĞĐŽǀĞƌŝŶŐ͕ƚŚĞ
ĐƵƌƌĞŶƚůŽĂĚĚĞĐƌĞĂƐĞƐĂŶĚŝƐĚĞƚĞƌŵŝŶĞĚĨƌŽŵsŶͲϭ

>ŽĂĚĨŽƌĞĂĐŚĞůĞŵĞŶƚŽĨ&ŵŽĚĞůĐĂůĐƵůĂƚĞĚĨƌŽŵYͼŐͼĚĞ ǁŚĞƌĞĚĞ ŝƐ

ƚŚĞĚĞƉƚŚŽĨǁĂƚĞƌŽǀĞƌĞĂĐŚĨŝŶŝƚĞĞůĞŵĞŶƚ

^ƚĂƚŝĐĂŶĂůLJƐŝƐƉƌŽĐĞĞĚƐƵƐŝŶŐĚLJŶĂŵŝĐƌĞůĂdžƚŝŽŶƐŽůǀĞƌ͕ĞŝƚŚĞƌƚŽĨƵůů
ĞƋƵŝůůďƌŝƵŵĐŽŶĚŝƚŝŽŶĨŽƌĐƵƌƌĞŶƚůŽĂĚĞĚƐƚĂƚĞŽƌĨŽƌƉƌĞĚĞĨŝŶĞĚƐĞƚŽĨ
ŝŶĐƌĞŵĞŶƚƐ͘
ZĞǀŝƐĞĚŐĞŽŵĞƚƌLJĂƚĞŶĚŽĨƐƚĂƚŝĐƌƵŶĚĞĨŝŶĞĚ͘

Fig. 5. Typical Pondr sequence for time stepping ponding analysis

318 Adam Bown et al. / Procedia Engineering 155 (2016) 313 – 322

The above flow chart illustrates the typical run sequence. In practical terms the geometric assessment is
undertaken from the FE mesh of the current model using RhinoTools. The revised hydrostatic loading in terms of
pond depth and area is calculated, and the salient inTENS input file updated. Each static inTENS run is then
automatically launched via the Windows command line.

3.3. Typical Results

The above process was applied to an FE model of a typical pneumatic air hall, starting internal pressure of 350Pa,
with a notional seed event of a 400kg weight applied to a membrane surface.

Fig. 6. FE model of typical air hall used for ponding assessment, initial depression introduced with weight

3.3.1. Structure Equilibrium

Further to the introduction of the seed event consider the case of a rain event occurring - the depression fills with
water. The ‘pond’ deflects the air hall downwards, the pond catchment area, volume and hydrostatic head datum
being corrected accordingly at each step. As a result of the structure deflection the ‘pond’ changes shape in plan,
some rainwater is able to flow over the structure edge and ultimately an equilibrium point is reached where
hydrostatic load is balanced by internal inflation pressure and does not increase in size. The structure is stable.

Fig. 7. Case of pond stabilization.

 Adam Bown et al. / Procedia Engineering 155 (2016) 313 – 322 319

3.3.2. Pond Recovery

Starting from the above stable state, now consider the above case of the internal pressure being increased in an
attempt to remove the pond. The air hall deflects, the pond initially flattens out, the rain water spreads but the
increase in pressure is sufficient to invert the pond and drain the rainwater over the structure side.

Fig. 8. Case of pond recovery further to pressure increase.

3.3.3. Pond Migration.

The above examples have been for the case of a symmetric depression. For the case of an asymmetric initial
depression, owing to the lack of lateral stability of the air hall, the pond rapidly migrates sideways until it ultimately
allows the pond to drop off the side of the structure. The applied load is lost and the air hall recovers to its static
inflated state. The air hall can be seen to be partially self-regulating.

Fig. 9. Pond migration.

320 Adam Bown et al. / Procedia Engineering 155 (2016) 313 – 322

With the pond moving away from the intial seed location, this example shows the importance of identifying and
tracking the pond low point and new catchment area at each step, as well as the hydrostatic datum.

3.4. Conclusions

With reference to examples undertaken on a pneumatic air hall, this section has demonstrated a potential
extension to regular hydrostatic load analysis for cases where the pond load varies in both magnitude and location in
space. The procedure is essentially a series of static analysis runs combined together using automated batch
processing. The geometric interrogation of the FE model is undertaken at the end of each static step, and the
geometric data used to update the pond catchment area, total area, depth and total load for the analysis input files.

4. Failure Propagation Collapse.

4.1. Introduction

Element failure is of interest to the tension structures engineer particularly when failure of the element in
question has the potential to lead to a progressive collapse. Whilst this may be manually analysed in the standard
static analysis environment (i.e. assessment of results, adjustment of model, re-analysis, continue) a batch processed
stepping tool, Tensys Tumblr, has been developed to automate the process.

The process defined in this paper will be limited to elements which may exhibit absolute failure such as cables
and woven fabrics, rather than elements which experience plastic behaviour prior to failure such as steel beams and
ETFE films.

4.2. Failure Value

A pre-requisite for such a failure analysis is the grouping of similar elements (cables and membrane) and
assignation of a failure load. The choice of value for failure load is often not immediately obvious. For example
structural cables will have an associated characteristic breaking strength and supplier defined limit tension. Woven
fabrics will have a documented breaking strength which may reduce over time. The choice of value for breaking
limit for each element lies with the expertise and experience of the engineer. Sequences should be run with varying
breaking values in order to determine the sensitivity of the structure.

4.3. Seed Event

Assuming the structure has been suitably analysed and elements sized to withstand the expected applied loads, a
‘seed’ event is required to initiate a failure event. This may be
• Extreme weather event
• Accidental damage
• Human interference, e.g. vandalism
The seed event may be represented by defining a particularly low failure value for one element within the
structure or alternatively simply removing a key element in the FE model. Alternatively the sequence defined below
maybe used to assess the response of a structure under applied load in a partially installed state.

4.4. Sequence

The sequence illustrated below is executed using a custom command line tool. When model adjustment is
required the ASCII input files are edited with the RhinoTools model interrogation software.
 Adam Bown et al. / Procedia Engineering 155 (2016) 313 – 322 321

&ŵŽĚĞůŽĨƐƚƌƵĐƚƵƌĞĚĞǀĞůŽƉĞĚĂŶĚƐƚĂƚŝĐĞƋƵŝůŝďƌŝƵŵĚĞĨŝŶĞĚ͘
&ĂŝůƵƌĞůŽĂĚŽĨŬĞLJĞůĞŵĞŶƚƐĚĞĨŝŶĞĚĂƐŝŶƉƵƚ͕&ĂůůŽǁĂďůĞ

^ĞĞĚĞǀĞŶƚĚĞĨŝŶĞĚ

^ƚĞƉϭ͗ŶǀŝƌŶŽŵĞŶƚĂůůŽĂĚĐĂƐĞĚĞĨŝŶĞĚƵƐŝŶŐ^//ŝŶƉƵƚĨŝůĞ͘^ƚĂƚŝĐĂŶĂůLJƐŝƐĐĂƐĞƵŶĚĞƌƚĂŬĞŶ͘
^ĞƚŽĨƚĞŶƐŝŽŶƐͬƐƚƌĞƐƐĞƐĚĞĨŝŶĞĚĨŽƌĞĂĐŚĞůĞŵĞŶƚ

^ƚĞƉϮ͗ZĞƐƵůƚĂŶƚĨŽƌĐĞƐ&ĐŽŵƉĂƌĞĚƚŽĂůůŽǁĂďůĞǀĂůƵĞƐĨŽƌĞĂĐŚĞůĞŵĞŶƚΖŶΖ

&ŽƌĐĂƐĞƐǁŚĞƌĞ&Ŷ х&ĂůůŽǁĂďůĞ͕ƐƚŝĨĨŶĞƐƐŽĨĞůĞŵĞŶƚΖŶΖƌĞĚƵĐĞĚƚŽnjĞƌŽŝŶ^//ŝŶƉƵƚĨŝůĞ
ŝ͘Ğ͘ƉůĂLJƐŶŽĨƵƌƚŚĞƌƉĂƌƚŝŶĂŶĂůLJƐŝƐ

^ĞƋƵĞŶĐĞĐŽŶƚŝŶƵĞƐƵŶƚŝůƐƚƌƵĐƚƵƌĞƐƚĂďŝůŝƐĞƐŽƌĐĂƚĂƐƚƌŽƉŝĐĐŽůůĂƉƐĞŽĐĐƵƌƐ

Fig. 10. Flow chart for sequential failure propagation analysis.

4.5. Typical Results

The above process was tested on an FE model of a typical tension structure with cable supports. A notional side
wind load was applied to the structure yielding the normal cable tension response as illustrated below.

WIND
DIRECTION

Fig. 11. Typical membrane and cable structure subjected to side wind load and resultant cable tensions

In order to initiate a progressive sequence one of the mast support cables was removed from the FE model and
the wind load re-applied. This results in a re-distribution of forces and subsequent sequence of failures, ultimately
terminating in the failure of one of the membrane panels. Key stages from this sequence are illustrated in figures 13
through 16 below.
322 Adam Bown et al. / Procedia Engineering 155 (2016) 313 – 322

Fig. 12. Seed event defined as failure (hence removal) of mast support cable. Tension under load increases in adjacent tie cable

Fig. 13. Sequence of migration of load around the structure. Adjacent elements in turn reach capacity and are automatically removed.

Fig. 14. As load re-distribution continues, stresses in membrane increase until ultimately membrane fails and is removed from the model.

4.6. Conclusions

This section provides a second example of the use of automated batch processing, to assess the potential for a
failure propagation of a combined membrane and cable structure. Whilst the tools to undertake this analysis are
present in many existing codes, batch processing coupled with a library of allowable forces allows engineers to
quickly examine larger numbers of potential failure sequences allowing sensitivity analysis.

References

[1] D.S. Wakefield, “Engineering Analysis of Tension Structures : Theory and Practice”, Engineering Structures ,Vol 21, pp.680-690, 1999
[2] D.S.Wakefield, “Membrane Engineering: Principles and Applications”, IASS Asia Pacific Conference on Shell and Spatial Structures,
Beijing, 1996
[3] Chertock, A., Cui, S., Kurganov, A. and Wu, T., 2015. Well balanced positivity preserving central upwind scheme for the shallow water
system with friction terms. International Journal for numerical methods in fluids, 78(6), pp.355-383